If the working voltage is 48V, I strongly recommend choosing ceramic capacitors with a withstand voltage of 100V, because their capacitance may drop significantly
I don't understand why to do this in such roundabout way.
If you need certain amount of actual C, look up the bias curves. Assuming a 100V rated part would deliver more C than a 50V rated part has statistical chances to go right, but it's still just a guess. If they are in the same package (height included) and same rated capacitance, difference in actual capacitance at 50V might be small or none.
Volumetric energy density (i.e., package size given the same C) is already better indicator than rated voltage, but even that is iffy.
Only if bias curves are not available and you can't afford buying parts from manufacturers who publish them (like TDK), then you need to guess, or measure.
In any case, Simon, concentrate on the following:
* Make sure you get enough actual C under your 48V bias. Try to find parts with this information available, no guessing.
* For large-C DC links, often it's a good idea to calculate uF_actual per dollar or per area or whatever. It may happen a part rated to smaller capacitance offers more bang for buck! Quickly done in Excel.
* Especially if parts are bigger than 0805, pay extreme care to avoid any board flex.
* Consider soft terminated parts, if you can afford them. For high reliability, it's much better investment than extra voltage rating.
* Remember, ceramics fail short, so if the upstream power supply can supply more than >10W, consider each capacitor as ignition source!
* Soft terminated parts greatly reduce this risk, but it's still not zero. Make sure there are no combustible materials nearby.
* Finally, if all these check out, 50V rated part is just fine on 48V bus. MLCCs can be used over the max voltage rating, but obviously managers do not like that, so keep within ratings. But no derating is needed.